MODELING DIMENSIONS IN THE XDW MODEL

A LVM-Driven Approach

R. Rajugan, Elizabeth Chang and Tharam S. Dillon

Digital Ecosystems and Business Intelligence Institute (DEBII), Curtin Univeristy of Technology, Australia

Keywords: XML, data/document warehouse, views, Object-Oriented conceptual models, Layered View Model (LVM),

XML views.

Abstract: Since the introduction of eXtensible Markup Language (XML), XML repositories have gained a foothold in

many global (and government) organizations, where, e-Commerce and e-Business models have maturated in

handling daily transactional data among heterogeneous information systems. Due to this, the amount of data

available for enterprise decision-making process is increasing exponentially and are being stored and/or

communicated in XML. This presents an interesting challenge to investigate models, frameworks and

techniques for organizing and analysing such voluminous, yet distributed XML documents for business

intelligence in the form of XML warehouse repositories and XML marts. In our previous work, we

proposed a Layered View Model (LVM) driven, conceptual modelling framework for the design and

development of an XML Document Warehouse (XDW) model with emphasis on conceptual and logical

semantics. There, we presented a view-driven framework to conceptually model and deploy meaningful

XML FACT repositories in the XDW model. Here, in this paper, we look at the hierarchical dimensions and

their theoretical semantics used to design, specify and define dimensions over an XML FACT repository in

the XDW model. One of the unique properties of this LVM-driven approach is that the dimensions are

considered as first-class citizens of the XDW conceptual model. Also, here, to illustrate our concepts, we

use a real-world case study example; a logically grouped, geographically dispersed, XDW model in the

context of a global logistics and cold-storage company.

1 INTRODUCTION

Data Warehousing (DW) has been an approach

adopted for handling large volumes of historical data

for detailed analysis and management support.

Transactional data in different databases is cleaned,

aligned and combined to produce good data

warehouses. At the most basic level, data

warehousing has been an approach adopted for

management of large volumes of historical data for

detailed analysis to provide crucial business

intelligence (BI) for organisations in: (i) Decision

Support Systems (DSS) (Elmasri & Navathe 2004;

Gray & Watson 1998), (ii) Management Information

Systems (MIS) (Elmasri & Navathe 2004) and (iii)

Executive Information Systems (Gray & Watson

1998).

A data warehouse integrates large amounts of

enterprise data from multiple and independent data

sources consisting of operational databases into a

common repository (Feng & Dillon 2003) for

querying and analysis (using BI tools). In addition,

data warehouses are designed for online analytical

processing (OLAP) (Elmasri & Navathe 2004; Feng

& Dillon 2003; Kimball & Ross 2002; Trujillo,

Luján-Mora & Song 2003), where the queries

aggregate large volumes of data in order to detect

trends and anomalies. To reduce the cost of

executing aggregate queries in such an environment,

warehousing systems usually pre-compute

frequently used aggregates and store each

materialized aggregate view (Feng & Dillon 2003;

Gopalkrishnan, Li & Karlapalem 1999; Theodoratos

& Sellis 1999) in a multidimensional data cube

(Feng & Dillon 2003; Gopalkrishnan, Li &

Karlapalem 1999; Gupta, Mumick & (eds) 1999;

Trujillo, Luján-Mora & Song 2003). These data

cubes group the base data along various dimensions,

corresponding to different sets of operational

attributes, and compute different aggregate functions

(e.g. sum, avg, min, max) on measures.

Rajugan R., Chang E. and S. Dillon T. (2007).

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach.

In Proceedings of the Ninth International Conference on Enterprise Information Systems - DISI, pages 60-70

DOI: 10.5220/0002372700600070

 SciTePress

In traditional data warehouse terminology, the

dimensional model is represented using FACTs and

dimensions. A FACT is a business performance

measurement usually numeric in nature and a

dimension refers to an independent entity that serves

as an entry point and/or mechanism to extract

meaningful measurements form the associated

FACT (Elmasri & Navathe 2004; Kimball & Caserta

2004; Kimball & Ross 2002). Also, depending on

the dimensional model (i.e. OO (Giovinazzo 2000),

Star Schema, O-R star (Mohammed 2001)) further

terminologies (e.g. FACT dimensions, shared

dimensions, etc.) are defined to elaborate some of

additional features of those models. In the relational

model, the popular dimensional model is the Kimbal

et al. Star Schema model (Kimball & Ross 2002),

where the FACT and the dimensions are represented

using tables (and materialized views), referred to as

FACT and dimension tables. A FACT table is a non-

normalized table with the numeric performance

measurements characterized by a group of (foreign)

keys drawn from the dimensional tables that form a

composite (foreign key).

Since its introduction in 1996, eXtensible

Markup Language (XML) (W3C-XML 2004) has

become the defacto standard for storing and

manipulating self-describing information (meta-

data), which creates vocabularies in assisting

information exchange between heterogenous data

sources over the web (Pokorn'y 2002). Due to this,

there is considerable work to be achieved in order to

allow electronic document handling, electronic

storage, retrieval and exchange. It is envisaged that

XML will also be used for logically encoding

documents for many domains. Hence, it is likely that

a large number of XML documents will populate the

would-be repository and several disparate

transactional databases. Conversely, Enterprise

Content Management (ECM) is the integration and

utilization of one or more technologies, tools, and

methods to capture, manage, store and deliver

content across an enterprise (ECM-AIIM 2005),

where XML is gaining momentum as the data

representation and integration language. One of the

data intensive issues in ECM is the data

warehousing concept that has gained in importance

in recent years (Elmasri & Navathe 2004).

The concern of managing large amounts of XML

document data raises the need to explore the data

warehouse approach through the use of XML

document marts and XML document warehouses.

Since the introduction of dimensional modelling

which, evolves around facts and dimensions, several

design techniques have been proposed to capture

multidimensional data (MD) at the conceptual level.

Ralph Kimball’s Star Schema (Kimball & Ross

2002) proved most popular, from which well-known

conceptual models SnowFlake and StarFlake were

derived. More recent comprehensive data warehouse

design models are built using Object-Oriented

concepts (structural relationships, Object cubes or

data cubes) on the foundation of Star Schema. In

(Abelló, Samos & Saltor 2001; Lujan-Mora, Trujillo

& Song 2002a, 2002b; Luján-Mora, Trujillo &

Vassiliadis 2004; Trujillo et al. 2001) two different

OO modelling approaches are demonstrated where a

data cube is transformed into an OO model

integrating class hierarchies. The Object-Relational

Star schema (O-R Star) model (Mohammed 2001;

Rahayu et al. 2001) aims to envisage data models

and their object features, focusing on hierarchical

dimension presentation, differentiation and their

various sorts of embedded hierarchies.

These models, both object and relational, have a

number of drawbacks, namely: data-oriented without

sufficient emphasis or capturing user requirements;

extensions of semantically poor relational models

(Star, Snowflake models) (Giovinazzo 2000; Rahayu

et al. 2001), original conceptual semantics are lost

before building data warehouses as the operational

data source is relational, further loss of semantics

resulting from oversimplified dimensional modelling

(Nunamaker, Chen & Purdin 1991; Rahayu et al.

2001), time consuming if additional data semantics

are required to satisfy evolving user requirements,

and complex query design and processing is needed,

therefore maintenance is troublesome (Inmon,

Imhoff & Battas 1996; Mohania, Karlapalem &

Kambayashi 1999). In applying these approaches to

the design of XML document warehouses, it is

important to consider XML’s non-scalar, set-based

and semi-structured nature. Traditional design

models lack the ability to utilise or represent XML

design level constructs in a well-defined abstract and

implementation-independent form.

Thus, to resolve some of these issues, in our

work, we consider a conceptual modelling (that

includes user requirement modelling (Vicky Nassis

et al. 2006) approach in proposing a document

warehouse model for XML. In the following

sections we present this model and the associated

semantics in detail.

This paper is organised as follows: Section 2

provide the motivation behind the XDW model

proposal, followed by a brief description our

research in Section 3. Section 4 provides XDW

model semantics in detail, followed by the

description of the illustrative real-world case study

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach

example used in this paper. Section 6 concludes this

paper with some discussion on our future research

directions.

2 MOTIVATION

Only recently, data warehouse models have focused

on incorporating conceptual semantics and user

requirements as part of the model specification. This

kind of work is still in the early stages; only a few

such works consider conceptual semantics (in

contrast to operational data oriented classical data

warehouse designs such as the Star Schema

(Kimball & Ross 2002) model) and are focused on

both conceptual and user requirements as part of the

DW design process.

In this paper, we present our approach to this

problem: a view-driven conceptual framework for

developing dimensional conceptual models for XML

documents.

Our work is radically different from existing

works such as (Gopalkrishnan, Li & Karlapalem

1999; Jeong & Hsu 2001; Lucie-Xyleme 2001;

Luján-Mora, Trujillo & Vassiliadis 2004; Luján-

Mora, Vassiliadis & Trujillo 2004; Medina, Luján-

Mora & Trujillo 2002; Mohammed 2001; Mohania,

Karlapalem & Kambayashi 1999). This is because,

in these DW models, views are mainly used to

provide aggregate data and queries, performance (as

materialized views), meta-data and OLAP queries

(Gupta, Mumick & (eds) 1999; Humphries, Hawkins

& Dy 1999; Mohania, Karlapalem & Kambayashi

1999; Trujillo, Luján-Mora & Song 2003). Little

work has been done in the direction of using views

for providing DW architectural constructs and

frameworks (Gopalkrishnan, Li & Karlapalem 1999;

Theodoratos & Sellis 1999).

For XML data, one of the early XML data

warehouse implementations for web data includes

the Xyleme Project (Lucie-Xyleme 2001). The

Xyleme project (Xyleme 2001) was successful and it

was made into a commercial product. It has well

defined implementation architecture and proven

techniques (such as materialised views) to collect

and archive web XML documents into an XML

warehouse for further analysis. Another approach by

Fankhauser et al. (Fankhauser & Klement 2003)

explores some of the changes and challenges of a

document-centric XML warehouse. Other works that

use XML in a data warehouse context include

(Golfarelli, Rizzi & Vrdoljak 2001; Medina, Luján-

Mora & Trujillo 2002). Our research is different

from these works as views are used to model and

design dimensional data instead of using views for

the purpose of providing data granularity,

dimensional refinements and/or for performance

(e.g. materialized views). The XDW is designed for

XML data and documents by incorporating XML

specific data semantics.

Our work is also different from approaches such

as OMG’s CWM approaches (OMG-CWM 2001),

where metadata model specifications based on MOF

(OMG-MOF™ 2003) were proposed for developing

a (mostly relational) data warehouse conceptual

model. Conversely, the works such (Lujan-Mora,

Trujillo & Song 2002a, 2002b; Luján-Mora, Trujillo

& Vassiliadis 2004; Luján-Mora, Vassiliadis &

Trujillo 2004; Trujillo, Luján-Mora & Song 2003;

Trujillo et al. 2001), where conceptual models of a

(mostly relational) data warehouse are developed

using Object-Oriented (OO) techniques and

languages were proposed. Though they are similar to

our work from a conceptual modelling point of view,

the works do not include: (a) an architectural

framework to develop a common framework for

different data domains; (b) explicit data warehouse

requirement specification and notational

representation of such user requirements; and (c)

constructs to model semi-structured (e.g. XML) data

model specific semantics such as ordering. But in

the work, some new directions have been proposed

to support OO concepts in the traditional FACT

driven data warehouse models.

As stated before, in this research, we look at

utilizing views in the Layered View Model (LVM)

for XML (R.Rajugan 2006; R.Rajugan et al. 2005)

as the foundation for developing conceptual

framework for dimensional modelling rather than

representing aggregate and/or dimensional queries

(and query wrappers). Also, the design of the

dimensional model is focused on capturing and

modelling user requirements (Vicky Nassis et al.

2006; Vicky Nassis et al. 2006), as opposed to

developing dimensional models using available

operational data and the associated semantics

(Kimball & Caserta 2004; Luján-Mora & Trujillo

2004). In summary, the main motivation for XDW

research includes:

(i) User requirements: separation of

operational data semantics and data

warehouse user requirements in the case of

XML data and document,

(ii) Top-down approach: separation of

implementation concerns (data format,

structure, etc.) from (XML) data warehouse

conceptual models,

ICEIS 2007 - International Conference on Enterprise Information Systems

(iii) Expressiveness: formulation of dimensional

semantics that are capable of expressively

modelling XML data (both data and

document centric) semantics,

(iv) XML: providing dimensional semantics that

can be expressed and described using XML

(and XML Schema) itself and

(v) Views in the LVM: investigating application

of the LVM for dimensional modelling in

achieving (ii) – (iv) above.

Another motivation is the design of DW using

conceptual semantics such as in OMG’s Model-

Driven Architecture (MDA) initiative (OMG-MDA

2003). Since the introduction of the MDA initiative,

platform independent models play a vital role in

system development and data engineering. Under the

MDA initiative, first the model of a system is

specified via an abstract notation independent of the

technical or deployment specifications (i.e. Platform

Independent Model or PIM), and then the PIM is

mapped or transformed into a deployment model

(i.e. Platform Specific Model or PSM) by adding

platform or deployment specific information into the

PIM. To support MDA initiatives in ECM (i.e. data

engineering, data semantics, constraints etc.), model

requirements have to be specified precisely at a

higher level of abstraction. This presents an

opportunity to investigate conceptual views as a

means of providing data abstraction and semantics in

PIMs for data intensive MDA solutions.

It should be noted that, though we refer to XDW

as an XML Document Warehouse, the concepts

presented here are common to both data-centric and

document-centric XML documents.

3 OUR WORK:

LVM-DRIVEN XML

DOCUMENT WAREHOUSE

(XDW) MODEL

The XDW model proposed in this research is

composed of four design levels, namely:

(i) XDW requirements level

(ii) XDW conceptual level

(iii) XDW logical (or schema) level

(iv) XDW document (or instance) level

Here, except for the requirements level, the other

three levels are analogous to the layers of abstraction

in the LVM. The XDW requirements level, in

addition to the layers of abstraction, enforces user

requirements in the form of (XML) Warehouse

Requirements (WR) and User Requirement (UR)

(Vicky Nassis et al. 2006; Nassis et al. 2005a),

which complements the XDW conceptual model. A

context diagram of this model is given in Fig. 1.

Thus, the uniqueness of the XDW model is also in

its approach to capturing and specifying data

warehouse requirements. This is because,

traditionally, a data warehouse model is heavily

constrained by the available operational data and its

structure, as warehouse modellers and designers

design a data warehouse using bottom-up

approaches (or reverse engineering warehouse

requirements), working from operational data to the

warehouse conceptual model.

Thus, it should be noted here that, in comparison

with traditional data warehouse requirements, as

unique to XDW design, we first develop the UR

model using specialized notations that are

independent of the operational data and/or data

structures. Also, the UR model is developed first

before constructing the conceptual model of the

XDW, but it is iteratively validated against the

available operational data, (conceptual) model

and/or structures. By adopting this approach, we

intend to model and represent user requirements that

are valid yet independent of the operational data.

Also, in addition to adopting user requirement

driven XDW design, the XDW model outlined

below, to best of our knowledge, is unique in its

kind as it is utilizes XML itself (together with XML

Schema) to provide; (i) structural constructs, (ii)

metadata, (iii) validity, and (iv) expressiveness (via

refined granularity and class decompositions).

Figure 1: The XDW Model (context diagram).

As shown in Fig. 1, the first design level is the

user requirement level which has two components,

namely: (a) warehouse requirements, and (b) user

requirements.

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach

The second design level (Fig. 1) is the XDW

conceptual model that has two main components,

namely: (a) the XML FACT Repository (xFACT)

and (b) the Virtual Dimensions (VDim).

The third level is the logical model of the XDW,

where the schemata transformation of the xFACT

and the associated VDims to (XML) schemas

occurs. The fourth level is the transformation of

VDim construct (i.e. conceptual operators

(R.Rajugan 2006)) to document level query

expressions using one or more native or embedded

query languages (e.g. XQuery, SQL ‘03).

In this paper, we focus only on the conceptual

and the theoretical semantics of the VDims. xFACT

model is discussed in detail in (R.Rajugan, Chang &

Dillon 2005) and the modelling and transformation

of VDim from conceptual level to logical (and

instance level) transformation are analogous to the

views in the LVM, as discussed in detail in

(R.Rajugan et al. 2005, 2006). Thus, we do not

include these discussions in this paper.

3.1 XDW Conceptual Level

As stated earlier, the XDW conceptual model is

composed of: (a) an XML FACT repository

(xFACT); and (b) a collection of associated,

logically grouped conceptual views that satisfies one

or more user requirements given in the UR model.

The xFACT is a snapshot of the underlying

transactional system(s) for a given context.

As defined earlier in (R.Rajugan, Chang &

Dillon 2005; R.Rajugan et al. 2005), a context is

more than a measure or an item that is of interest for

the organization as a whole. In classical data

warehouse models, a context is normally modelled

as an ID packed FACT and associated data

perspectives as dimensions. Usually, due to

constraints of the relational model, a FACT will be

collapsed to a single table, with IDs of its

dimension(s), thus emulating (with combination of

one or more dimension(s)) a data cube (or

dimensional data). A complex set of queries are

needed to extract information from the FACT-

Dimension model. But, in regards to XML, a context

is more than a flattened FACT (or simply referred to

as meaningless FACT) with embedded semantics

and constraints. It will also have embedded

relationships such as those featured in OO models

and semi-structured data such ordered composition,

exclusive disjunction etc. in addition to non-

relational constructs such as set, list, and bag.

Therefore, we argue that, a FACT structure similar

to a (relational) FACT table without the required

semantics does not provide semantic constructs that

are needed to accommodate an XML context.

The role of conceptual views is to provide

perspectives to the document hierarchy stored in the

xFACT repository. Since conceptual views can be

grouped into logical groups, each group is very

similar to that of a subject area (or class categories)

(Dillon & Tan 1993) in OO conceptual modelling

techniques. Each subject-area in the XDW model is

referred to as a cluster of Virtual Dimensions

(VDim) in accordance with dimensional models.

VDim is called virtual; that is, since it is modelled

using XML conceptual views (which are imaginary

XML documents) in the LVM and behaves as a

dimension for the given xFACT. In this paper we

only elaborate on VDims. A detailed discussion on

xFACT can be found in our work in (R.Rajugan,

Chang & Dillon 2005).

3.2 View-Driven Virtual Dimensions

A user requirement, which is captured and specified

in the XDW requirement model (namely UR model),

is transformed into one or more conceptual views in

the LVM, which are referred to as Virtual

Dimension/s, (VDim) in association with the

xFACT. These are typically views involving

aggregation or perspectives of the underlying

xFACT, which serves as the pre-defined context.

A valid user requirement is such that, it can be

satisfied by one or more conceptual views for a

given context (i.e. xFACT). But in the case where

for a given user requirement there is no transactional

document or data fragment to satisfy it, further

enhancements are necessary to make the

requirement feasible to model with a certain xFACT.

Therefore, modelling and specifying VDim is an

iterative process, where user requirements are

validated against the xFACT in conjunction with the

operational data and data semantics. Thus, VDim is

an additional elaboration, extraction and/or

specification of the required information from the

xFACT (thus, from the aggregated operational data).

It should be noted here that, conceptual views are

first-class citizens of the conceptual model.

Therefore, since a VDim is a conceptual view,

VDim is also a first-class citizen of the conceptual

model.

VDim can be materialized (for data refinement

or for the purpose of performance issues such as

relational views in classical Star model) or

aggregated further by defining additional conceptual

views to refine and/or satisfy further user

requirements.

ICEIS 2007 - International Conference on Enterprise Information Systems

4 XDW MODEL SEMANTICS

In this section, we present some of the formal

semantics associated with the XDW model (without

the UR model). Since the proposed model is driven

by views of LVM, some of the concepts and

definitions are extensions and/or an elaboration of

the concepts and definitions presented in (R.Rajugan

et al. 2005).

The XDW conceptual model consists of an

xFACT repository and multiple hierarchical

dimensions. Thus, at first glance, the XDW is

analogous to the Star/Snowflake schema

(Gopalkrishnan, Li & Karlapalem 1999) of the

relational model, or the Operational Data Store

(ODS) (Inmon, Imhoff & Battas 1996) model,

except that both the xFACT and the VDims are

modelled using views in the LVM. Also, since

xFACT is more complex than a relational FACT

table, it can be considered as one context (e.g. sales)

that is of interest to the organization, with multiple

sub-contexts (such as regional-sales, sales-by-city,

sales-by-store etc.). Therefore it can be shown that

there exists a many-to-many (m:n) relationship

between one xFACT and a VDim (or a VDim

hierarchy).

4.1 xFACT and View-Driven VDim

Let XML FACT repository (xFACT) be denoted

FACT

x and a virtual dimension (VDim) denoted as

Dim

V in the XDW model. In work with LVM for

XML, we defined a conceptual view using a context.

Since each VDim is a conceptual view, by

definition, a virtual dimension

Dim

V can be defined

as:

Definition 1. A virtual dimension is a conceptual

view

Dim

V , such that

Dim

is a 4-ary tuple of

Dim

rel

Dim

obj

Dim

name

VVV ,, and

Dim

constra

int

, where

Dim

name

V is the

name of the virtual dimension

Dim

V ,

Dim

obj

V is a set

of objects in

Dim

V ,

Dim

rel

is a set of object

relationships in

Dim

V , and

Dim

constra

int

is a set of

constraints associated with

Dim

obj

V and

Dim

rel

Dim

V .

),,,(

int

Dim

constra

Dim

rel

Dim

obj

Dim

name

Dim

VVVVV =

(1)

If one considers an xFACT to be a context, it can

be shown as in definition 2, below using the

definition of context presented in (R.Rajugan et al.

2005, 2006) as:

Definition 2. An XML FACT (xFACT)

repository

xF is defined such that, xF consists of

a xFACT name

name

xF , a set of objects

obj

xF ,, a set

of object relationships

rel

xF , and a set of

constraints associated with its objects and

relationships

intconstra

xF .

),,,(

intconstrarelobjname

xFxFxFxFxF =

(2)

Similar to the definition of a valid conceptual

view (R.Rajugan et al. 2005, 2006), here we can

define a valid virtual dimension as:

Definition 3. A virtual dimension

Dim

V called a

valid virtual dimension for a given XML FACT

(xFACT) repository

, if and only if for any

object

Dim

obj

Vobj ∈∀ , there exist objects

objn

xFobjobjobj ∈∃ ,...,,

, such that

),....,(....

11 nm

objobjobj

, where

∈

....

and

be a set of conceptual

operators. That is,

obj is a newly derived object

from existing objects

objobjobj ,....,,

via

a series of conceptual operators

....

From definition 3, it is intuitively deductible that,

xF for a given

Dim

V is actually the context for

the

Dim

V question.

4.2 XDW Relationships

Typically, in an XDW model, for one xFACT, there

exists one or more VDims. Let the total number of

Dim

V be n. Let

R denote the relationship

between the xFACT and a VDim and

R between

two dimensions. The relationship between xFACT

xF and VDim

Dim

V , may be denoted as:

),(

Dim

VxFR =

(3)

where

≤

0 .

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach

Also, the hierarchical relationship (dimensional

hierarchy) between two VDims,

Dim

V and

Dim

can be shown as:

),(

Dim

VVR

(4)

Both

R and

R may fall into one or more of

the dimension specific relationships type (and

constraints) as: (a) aggregate dimension (minimum ,

maximum, count, average), (b) time-variant

dimension, (c) subject-variant dimension and (d)

aggregate-descriptive dimension.

These types of relationships may correspond to

one or more of the dimensional (conceptual)

operators or queries, as described in (Nassis et al.

2005b). They may be grouped into: (i) aggregate

selection, (ii) aggregate sort/order, (iii)

implicit/explicit joins and (iv) aggregate grouping.

Analogous to the views in the LVM, at the

logical level, these dimensional (conceptual)

operators or queries are transformed into W3C use

case query context algorithms (Nassis et al. 2005b)

and later at the document/instance level to language

specific query expression, such as XQuery and/or

SQL.

5 ILLUSTRATIVE CASE STUDY

EXAMPLE

As an illustrative case study for this paper, we intent

to gradually build a e-Sol case study described in

(Chang et al. 2003; ITEC 2002; R.Rajugan et al.

2005) as a XDW conceptual and logical model for

the purposes of archiving and analysing e-Sol data

for the purpose of business planning and

intelligence. We refer to the e-Sol XDW model as e-

Sol-W. Given below is the extended description and

requirements of the e-Sol-W for the purpose of

building a data warehouse and data marts.

For e-Sol to support DSS, EIS and MIS, it is

essential to provide a data model to support

dimensional data in the context of a data warehouse.

Due to e-Sol’s dynamic and heterogeneous nature

(both system and data), the data warehouse model

should support rapidly evolving new data formats

(from relational, XML to propriety data scripts), at a

high level of abstraction. From a local

stakeholder/partners’ perspective, the XDW model

solves some of the problems faced by e-Sol. But

from a global perspective, where multiple

stakeholders/partner systems are involved (i.e.

collaborative partners, global customers, etc.) and

there is a need to support e-Sol’s global information

demand, the role and scope of XDW has to evolve

and a new global warehouse model is inevitable and

an unfortunate reality.

To illustrate our concepts, we highlight a few,

simplified XDW user requirements. We consider a

simplified XDW model for archiving and analysing

warehouse bookings, income and capacity for the

warehouses in the e-Sol. Some of these requirements

include (Fig. 2-3):

(i) Warehouse booking: Warehouse booking

records by (a) customers, (b) companies

and (c) collaborative partners, grouped by

(i) year, (ii) month and (iii) by warehouse

location or by region (e.g. Asia-Pacific,

China etc.). This information may help to

rate customers and to plan warehouse

capacity around the world, for a given time

of the year.

(ii) Warehouse Capacity: Sort warehouse usage

(i.e. slack space and near-full capacity

measure) by year, month and quarterly and

individual Q1, Q2, Q3 and Q4 capacity

measure for (a) warehouse by region, and

(b) warehouse by country.

(iii) Warehouse Revenue: List warehouse

revenue by (a) year, (b) month (c) quarterly

and (d) individual Q1, Q2, Q3 and Q4 for

(i) individual warehouses, and (b)

warehouses by region.

In our research, we use UML as the OO

modelling language to represent XDW conceptual

level artefacts (Fig. 2-3). Here, we use of UML just

as a modelling notation as it is easily understood and

standardised. As we have stated in our work in

LVM, other OO languages may also be used instead

of UML. It should be noted here that, we do not

discuss the modelling and design of the xFACT here

and the Fig. 3 is given only for illustrative purpose.

A detailed discussion on the modelling xFACT can

be found in (R.Rajugan, Chang & Dillon 2005).

Example 1: For example, as shown in Fig. 2, in

the e-Sol-W case study, a VDim hierarchy, where

there exist inheritance relationships between the

VDim

Quarterly_WarehouseCapacity and the

individual Q1, Q2, Q3 and Q4 warehouse capacities.

Here, to model and represent XDW concepts, we

utilize UML stereotypes. We introduced a new UML

stereotype called

<<VDim>> to model the virtual

dimensions at the XDW conceptual level. Analogous

to conceptual views in the LVM, this stereotype is

similar to a UML class notation with a defined set of

attributes and methods. The set of methods here can

ICEIS 2007 - International Conference on Enterprise Information Systems

have either: constructors (to construct a VDim) or

manipulators (to manipulate the VDim attribute set).

Similar to conceptual views, at the XDW conceptual

level, VDims can have additional semantic

relationships such as generalization, aggregation,

association and these can be shown using standard

UML notation. In addition to this, two VDims can

also have

<<construct>> relationships with

dependencies. Similarly, an xFACT can be

represented using the

<<xFACT>>, as shown Fig. 2-

We have stated that semantically related

conceptual views could be logically grouped

together as grouping of classes into a subject area.

Further, a new view-hierarchy and/or constructs can

be added to include additional semantics for a given

user requirement. In the XDW conceptual model,

when a collection of similar or related conceptual

views are logically grouped together, we called it

grouped VDims (Fig. 2), implying that it satisfies

one or more logically related user requirement(s).

In addition, we can also construct additional

conceptual view hierarchies as shown in Fig. 2.

These hierarchies may form additional structural or

dependency relationships with existing conceptual

views or view hierarchies (grouped VDims) as

shown in Fig. 2. Thus, it is possible that a cluster of

dimensional hierarchy/ies can be used to model a

certain set of user requirement/s. Therefore we argue

that, this aggregate aspect can give us enough

abstraction and flexibility to design a user-centred

XDW model.

In order to model a hierarchy of VDim and

capture the logical grouping among them, we utilize

the package construct in UML. Thus, a grouped

VDim hierarchy as shown in Fig. 2 can be

represented using the UML package notation. This

in practice describes our logical grouping of

conceptual views and their hierarchies. Thus, we

utilize packages to model our connected dimensions

(Fig. 2).

In Fig 2, we show our case study XDW model

with xFACT and VDims connected via the

<<construct>> stereotype. Also, following the

arguments presented, we can show that, the xFACT

(shown in Fig 2-3) can be grouped into one logical

construct and can be shown in UML as one package.

Example 2: In the e-Sol-W example, as shown

in Fig. 2,

WMS_Warehouse is an xFACT (Fig. 3)

represented using a stereotyped UML package.

Example 3: Also, in Fig 2, a VDim, xFACT, and

the relationship between VDim and the xFACT are

shown using

<<VDim>>, <<xFACT>> and

<<construct>> stereotypes.

Figure 2: VDim hierarchy in e-Sol-W case study (with UML stereotypes).

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach

Example 4: The VDim hierarchical relationships

(Fig. 2) are shown using OO part-of (composition)

relationships or using the

<<construct>>

stereotyped (dependency) relationships (i.e. view of

a view).

Example 5: For example, as shown in Fig. 2, in

the e-Sol-W case study, a VDim hierarchy, where

there exist a part-of relationships between the VDim

Quarterly_WarehouseCapacity and the

individual Q1, Q2, Q3 and Q4 warehouse capacities.

Example 6: In e-Sol-W example, a conceptual

views

Q1_Warehouse-Revenue, Q2_

Warehouse-Revenue

, etc. are constructed in the

given context of

Warehouse_Revenue (Fig 2). The

valid context is given by the e-Sol-W

WMS_Warehouse objects (Fig 3).

Example 7: Also in the e-Sol-W example (Fig.

2), a conceptual views

Q1_WarehouseCapacity,

Q2_WarehouseCapacity, etc. are constructed in

the given context of

Figure 3: An xFACT example in the e-Sol-W (WMS-Warehouse).

ICEIS 2007 - International Conference on Enterprise Information Systems

WarehouseCapacity_by_Year. Again the valid

context is given by the e-Sol-W

WMS_Warehouse

objects. Also, it should be noted that, due to the

nature of xFACT, further dimensions may be

constructed such as;

Regional-Warehouse-

Capacity-by-Season

, Warehouse-Capacity-

by-Country, etc. providing regional and/or global

perspectives.

Example 8: In Fig. 2, in the xFACT is stated as

a materialized at the conceptual level using OCL

like syntax (R.Rajugan 2006).

Example 9: Similar to the above example, in the

VDim hierarchy of,

WarehouseCapacity_by_Year is stated as a

materialized conceptual view (Fig. 2), implying that

it is a persistence view (or VDim) during the lifetime

of the system. VDim “

Warehouse-Revenue” is

also a materialized view.

6 CONCLUSION

In this paper, we presented an intuitive, a view-

driven, conceptual framework (similar to the PIMs

in MDA approach) to conceptually model, design

and implement dimensions using the XML FACT

repository in the XDW model.

For future work, some further issues deserve

investigation. First, the investigation into OLAP

support in the XDW model using such (virtual)

dimensions. Second is the providing formal

explanatory semantics to improve design and

deployment of such dimensions using formalisms

such as fuzzy sets. Finally, it is the formulation of a

valid empirical study to consider and validate

performance and quality issues using large datasets.

REFERENCES

Abelló, A, Samos, J & Saltor, F 2001, 'Understanding

facts in a multidimensional object-oriented model', 4th

Int. Workshop on DW and OLAP (DOLAP '01).

Chang, E, et al., T 2003, 'A Virtual Logistics Network and

an e-Hub as a Competitive Approach for Small to

Medium Size Companies', 2nd

Human.Society@Internet Conf., Korea, pp. 265-71.

Dillon, TS & Tan, PL 1993, Object-Oriented Conceptual

Modeling, Prentice Hall, Australia.

ECM-AIIM 2005, The ECM Association, AIIM,

http://www.aiim.org.

Elmasri, R & Navathe, S 2004, Fundamentals of database

systems, 4th edn, Pearson/Addison Wesley, New

York.

Fankhauser, P & Klement, T 2003, 'XML for Data

Warehousing Changes & Challenges', DaWaK 2003,

Springer, Prague, pp. 1-3.

Feng, L & Dillon, TS 2003, 'Using Fuzzy Linguistic

Representations to Provide Explanatory Semantics for

Data Warehouses', IEEE Trans. on Knowledge and

Data Engineering (TOKDE), vol. 15(1), pp. 86-102

Giovinazzo, WA 2000, Object-oriented data warehouse

design : building a star schema, Prentice Hall PTR,

Prentice-Hall Int., Upper Saddle River, N.J. London.

Golfarelli, M, Rizzi, S & Vrdoljak, B 2001, 'Data

warehouse design from XML sources', Proc. of the 4th

ACM Int. workshop on Data warehousing and OLAP,

ACM Press, Atlanta, Georgia, USA, pp. 40 - 7.

Gopalkrishnan, V, Li, Q & Karlapalem, K 1999,

'Star/Snow-flake Schema Driven Object-Relational

Data Warehouse Design and Query Processing

Strategies', (DaWaK '99), Springer, Florence Italy.

Gray, P & Watson, HJ 1998, Decision Support in The

Data Warehouse, Prentice Hall PTR, USA.

Gupta, A, Mumick, IS & (eds) 1999, Materialized views:

techniques, implementations, and applications, eds A

Gupta & IS Mumick, MIT Press.

Humphries, M, Hawkins, MW & Dy, MC 1999, Data

Warehousing: Architecture & Implementation,

Prentice Hall PTR, USA.

Inmon, WH, Imhoff, C & Battas, G 1996, Building the

operational data store, John Wiley & Sons, NY, USA.

ITEC 2002, iPower Logistics

(http://www.logistics.cbs.curtin.edu.au/), viewed 2005.

Jeong, E & Hsu, C-N 2001, 'Introduction of Integrated

View for XML Data with Heterogenous DTDs', Proc.

of the 10th Int. Conf. on Information and Knowledge

Management (CIKM '01), ACM, US, pp. 151-8.

Kimball, R & Caserta, J 2004, The data warehouse ETL

toolkit : practical techniques for extracting, cleaning,

conforming, and delivering data, Wiley, Hoboken, NJ.

Kimball, R & Ross, M 2002, The data warehouse toolkit :

the complete guide to dimensional modeling, 2nd edn,

Wiley, New York.

Lucie-Xyleme 2001, 'Xyleme: A Dynamic Warehouse for

XML Data of the Web', IDEAS '01, eds ME Adiba, C

Collet & BC Desai, IEEE Computer Society 2001,

Grenoble, France, pp. 3-7.

Luján-Mora, S & Trujillo, J 2004, 'A Data Warehouse

Engineering Process', Third ADVIS '04, Springer-

Verlag GmbH, Izmir, Turkey, p. 14.

Lujan-Mora, S, Trujillo, J & Song, I-Y 2002a, 'Extending

the UML for Multidimensional Modeling', Fifth Int.

Conf. on the Unified Modeling Language and its

applications (UML '02), Springer-Verlag London,

UK, Dresden, Germany, pp. 290-304.

Lujan-Mora, et al., 2002b, 'Multidimensional Modeling

with UML Package Diagrams', Proc. of the 21st Int.

Conf. on Conceptual Modeling (ER '02), Springer-

Verlag UK, pp. 199-213.

Luján-Mora, et al., 2004, 'Advantages of UML for

Multidimensional Modeling', Proc. of the 6th Int.

Conf. on ICEIS '04, Porto, Portugal, April.

MODELING DIMENSIONS IN THE XDW MODEL - A LVM-Driven Approach

Luján-Mora, S, Vassiliadis, P & Trujillo, J 2004, 'Data

Mapping Diagrams for Data Warehouse Design with

UML', 23rd Int. Conf. on Conceptual Modeling (ER

'04), vol. 3288, Shanghai, China, p. 191.

Medina, E, et al., J 2002, 'Handling Conceptual

Multidimensional Models Using XML through DTDs',

Proc. of the 19th British National Conf. on Databases:

Advances in Dbs, vol. 2405, Springer, pp. 66 - 9.

Mohammed, S 2001, 'Object-Relational Data Warehouse',

Master by Coursework (Major Thesis) thesis, La

Trobe University, Melbourne, Australia.

Mohania, MK, Karlapalem, K & Kambayashi, Y 1999,

'Data Warehouse Design and Maintenance through

View Normalization', 10th Int. DEXA '99, vol. 1677,

Springer, Florence, Italy, pp. 747-50.

Nassis, V, Dillon, TS, R.Rajugan & Rahayu, W 2006, 'An

XML Document Warehouse Model', The 11th Int.

Conf. on Database Systems for Advanced Applications

(DASFAA '06), Springer, Singapore, pp. 513-29.

Nassis, V, Dillon, TS, Rahayu, W & R.Rajugan 2006,

'Goal-Oriented Requirement Engineering for XML

Document Warehouses', in J Darmont & O Boussaid

(eds), Processing and Managing Complex Data for

Decision Support, Idea Group Publishing, pp. 28-62.

Nassis, V, R.Rajugan, Dillon, TS & Rahayu, JW 2005a, 'A

Requirement Engineering Approach for Designing

XML-View Driven, XML Document Warehouses',

COMPSAC '05, IEEE CS, Scotland, pp. 388-95.

Nassis, V, R.Rajugan, Dillon, TS & Rahayu, W 2005b,

'Conceptual and Systematic Design Approach for

XML Document Warehouses', Int. Journal of Data

Warehousing and Mining, vol. 1(3), no. 3, pp. 63-87

Nunamaker, JF, Chen, JM & Purdin, TDM 1991, 'System

Development in Information Systems Research',

Journal of Mgmt. of IS, vol. 7(3), pp. 89-106

OMG-CWM 2001, The Common Warehouse Metamodel

(http://www.omg.org/technology/cwm/), OMG, 2005.

OMG-MDA 2003, The Architecture of Choice for a

Changing World®, MDA Guide Version 1.0.1

(http://www.omg.org/mda/), OMG, 2005.

OMG-MOF™ 2003, Meta-Object Facility (MOF™), 1.4

(http://www.omg.org/technology/documents/modeling

_spec_catalog.htm#MOF), OMG, 2005.

Pokorn'y, J 2002, 'XML Data Warehouse: Modelling and

Querying', Proc. of the Baltic Conf. (BalticDB-IS '02).

R.Rajugan 2006, 'A Layered View Model for XML with

Conceptual and Logical Extension, and its

Applications', PhD thesis, University of Technology,

Sydney (UTS), Australia, Sydney.

R.Rajugan, Chang, E & Dillon, TS 2005, 'Conceptual

Design of an XML-View Driven, Global XML FACT

Repository for XML Document Warehouses', 1st Int.

Workshop on Data Management in Global Data

Repositories (GRep ‘05), held in conjunction DEXA

‘05

, IEEE CS, Copenhagen, Denmark, pp. 1139 - 44.

R.Rajugan, Chang, E, Dillon, TS & Feng, L 2005, 'A

Three-Layered XML View Model: A Practical

Approach', 24th Int. Conf. on Conceptual Modeling

(ER '05), Springer-Verlag, Austria, pp. 79-95.

R.Rajugan, et al., 2006, 'Modeling Views in the Layered

View Model for XML Using UML', Int. Journal of

Web Information Systems (IJWIS), Troubador

Publisher Ltd., vol. 2(2), pp. 95-117, June 2006.

Rahayu, W, Dillon, TS, Mohammed, S & Taniar, D 2001,

'Object-Relational Star Schemas', 13th IASTED Int.

PDCS '01, IASTED, LA, USA.

Theodoratos, D & Sellis, T 1999, 'Dynamic Data

Warehouse Design', 1st DaWak '99, Springer, Italy,

pp. 1-10.

Trujillo, J, et al., 2003, 'Applying UML For Designing

Multidimensional Databases And OLAP Applications.

13-36', Advanced Topics in Database Research, Idea

Group Publication, vol. 2, pp. 13-36

Trujillo, J, Palomar, M, Gomez, J & Song, I-Y 2001,

'Designing Data Warehouses with OO Conceptual

Models', IEEE Computer Society, "Computer",

December, 2001, pp. 66-75.

W3C-XML 2004, Extensible Markup Language (XML)

1.0, (http://www.w3.org/XML/), The World Wide Web

Consortium (W3C).

Xyleme 2001, Xyleme Project (http://www.xyleme.com/).

ICEIS 2007 - International Conference on Enterprise Information Systems